Python library to trace graphql calls with Datadog.
ddtrace-graphql
is tested with:
- Python versions: 3.5, 3.6, nightly
- graphql-core: 2.0, 1.1.0, latest
- ddtrace: 0.11.1, 0.10.1, latest
Screenshots for pyramid app serving GraphQL with tracing enabled:
$ pip install ddtrace-graphql
$ git clone https://github.com/beezz/ddtrace-graphql.git
$ cd ddtrace-graphql && python setup.py install
To trace all GraphQL requests patch the library. Put this snippet to your application main entry point.
__import__('ddtrace_graphql').patch()
# OR
from ddtrace_graphql import patch
patch()
Check out the datadog trace client for all supported libraries and frameworks.
Note
For the patching to work properly, patch
needs to be called
before any other imports of the graphql
function.
# app/__init__.py
__import__('ddtrace_graphql').patch()
# from that point all calls to graphql are traced
from graphql import graphql
result = graphql(schema, query)
Trace only certain calls with traced_graphql
function
from ddtrace_graphql import traced_graphql
traced_graphql(schema, query)
DDTRACE_GRAPHQL_SERVICE: | Define service name under which traces are shown in Datadog. Default value is graphql |
---|
$ export DDTRACE_GRAPHQL_SERVICE=foobar.graphql
Default arguments passed to the tracing context manager can be updated using
span_kwargs
argument of ddtrace_graphql.patch
or
ddtrace_graphql.traced_graphql
functions.
Default values:
name: | Wrapped resource name. Default graphql.graphql . |
---|---|
span_type: | Span type. Default graphql . |
service: | Service name. Defaults to DDTRACE_GRAPHQL_SERVICE environment variable if present, else graphql . |
resource: | Processed resource. Defaults to query / mutation signature. |
For more information visit ddtrace.Tracer.trace documentation.
from ddtrace_graphql import patch
patch(span_kwargs=dict(service='foo.graphql'))
from ddtrace_graphql import traced_graphql
traced_graphql(schema, query, span_kwargs=dict(resource='bar.resource'))
In case you want to postprocess trace span you may use span_callback
argument. span_callback
must be function with signature def callback(result=result, span=span)
where result
is graphql execution result or None
in case of fatal error and span is trace span object
(ddtrace.span.Span).
What is it good for? Unfortunately one cannot filter/alarm on span metrics resp. meta information even if those are numeric (why Datadog?) so you can use it to send metrics based on span, result attributes.
from datadog import statsd
from ddtrace_graphql import patch, CLIENT_ERROR, INVALID
def callback(result, span):
tags = ['resource:{}'.format(span.resource.replace(' ', '_'))]
statsd.increment('{}.request'.format(span.service), tags=tags)
if span.error:
statsd.increment('{}.error'.format(span.service), tags=tags)
elif span.get_metric(CLIENT_ERROR):
statsd.increment('{}.{}'.format(span.service, CLIENT_ERROR), tags=tags)
if span.get_metric(INVALID):
statsd.increment('{}.{}'.format(span.service, INVALID), tags=tags)
patch(span_callback=callback)
Some frameworks use exceptions to handle 404s etc. you may want to ignore some exceptions resp. not consider them server error. To do this you can supply ignore_exceptions argument as list of exception classes to ignore. ignore_exceptions will be used in python's isinstance thus you can ignore also using base classes.
from ddtrace_graphql import patch
patch(ignore_exceptions=(ObjectNotFound, PermissionsDenied))
from ddtrace_graphql import traced_graphql
traced_graphql(
schema, query,
ignore_exceptions=(ObjectNotFound, PermissionsDenied))
$ git clone https://github.com/beezz/ddtrace-graphql.git
$ pip install --editable ddtrace-graphql[test]
$ cd ddtrace-graphql
$ tox